Techno-Economic Solution
for Semi-Dispatchable Solar
P M D Nayanathara, H S Perera, R A H M P N N Abeyrathne, J R Lucas
Department of Electrical, Electronic and Telecommunication Engineering, General Sir John Kotelawala
Defence University, Sri Lanka
dulakshinayanthara@kdu.ac.lk
Abstract— Solar power generation is subject to the natural
unpredictability of local weather conditions, particularly
intermittency of cloud cover. If solar generation is directly
coupled to the grid, system control would have little
control of power to the system. Therefore, a significant
solar capacity installed, can have potentially a negative
impact. The paper presents the use of an energy storage
device, coupled with a forecasting mechanism, to convert
solar into a more predictable, semi-dispatchable input. The
stochastic nature of the solar output is characterized and
fed into the power system in half-hourly profiles obtained
utilizing weather forecast data and past trends. The
random deviations that would not be categorized by the
prediction are mitigated using the energy storage device
which assists in adhering to the pre-scheduled infeed,
dependent on predicted weather attributes. The
forecasting mechanism comprises of solar irradiation,
recent energy trends and five weather attributes: humidity,
cloud cover, precipitation, visibility and gusts. Results show
that an accuracy of 88% could be achieved for the half-
hourly predictions. The use of a battery of 15% allows the
predicted hourly energy to be quite accurately fed during
the day, giving around 93% for the daily energy prediction.
Keywords— solar intermittency, battery storage, semi-
dispatchable, forecasting
I. INTRODUCTION
In compliance with the “National Energy Policy and
Strategies”, the “Generation Expansion Plan of the Ceylon
Electricity Board (CEB) for 2018-2037” [1] focuses on the
growth of solar energy sector. Relevant strategies:
increasing the share allocated to renewable energy
resources, reducing energy losses in transmission and
distribution systems, minimization of Green House Gas
(GHG) emission by the energy sector.
In order to support this mission, the expansion plan has
envisaged an addition of 1232 MW of solar power
generation, especially decentralized rooftop solar plants in
Sri Lanka [1]. Figure 1 illustrates the wide varying solar
infeed for the week 8th to the 14th October 2018 together
with an almost perfect day (21st October 2018) which
shows that during the same week, the level of output can
vary within the range of 90% to 20%.
This unpredictable and intermittent nature of solar power
is a drawback on the interest for solar from both the
power producer and utility provider.
Figure 1. Weekly Solar Output
Constant fluctuations caused by intermittency make it
even more unpredictable. Hence, power output of the
solar plant is not controllable as it remains dependent
upon irradiation from the sun, which is affected by a
number of stochastic variables. Seasonal nature of
irradiation and monsoon periods amplify this situation. In
addition to this, solar plants also lack inertia and reduces
the controllability further down.
These inherent properties of solar power results in
providing a lower value of service. As a solution, CEB
recommends a battery bank to store the energy produced
during daylight hours enabling the energy to be utilized
later.
However, power producers hesitate to adopt this method
due to cost concerns. Therefore, a more approachable and
a cost-effective method, which only requires storing of a
minor portion of the energy available in the daytime, is
presented. Reduced initial cost encourages the power
producers to implement the proposed system.
Said method aims to convert non-dispatchable solar
power into semi-dispatchable condition with the aid of
forecasting and storage. It develops a forecasting model to
predict the solar power output sufficiently in advance.
Then informs the prediction to CEB System Control Center
to take necessary action. A battery storage, equipped with
an automatic switching mechanism, is implemented to
maintain the predicted power output, irrespective of solar
intermittency. Proposed enhanced and value-added solar
output should be rewarded under a semi-dispatchable
tariff policy.
 II. METHODOLOGY Lanka only experiences minute seasonal changes, a
A. Forecasting monthly classification was considered more suitable [11].
1) Weather based Solar Forecasting Next data classification was based on time of the day. For
A roof-top solar panel installation, of 4.16kW located in this, the timeline for a day was segmented into several
Mount Lavinia for an area of 26 m2, was selected for the time bands and the resulting correlations with solar
study. Daily energy readings and instantaneous power production were tested. The forecasting model was built
readings at fifteen-minute intervals are available. Weather for the timeline segmentation that gave the best
data required to develop the forecasting system are correlation with the solar production. It is indicated in
obtained from AccuWeather.com. AccuWeather was Table 1.
chosen as the weather source following a detailed study Table 1. Timeline Classification
cum analysis of data availability, relevancy and reliability
Time band Number Allocated Time
from a number of weather sources.
Forecasting systems, tried utilizing existing formula from TB00 0000 to 0730 hours
literature for the prediction of the expected solar TB01 0730 to 1000 hours
irradiation [5] [6], failed to predict the intermittent short TB02 1000 to 1430 hours
fluctuations. Precise study of data patterns reveal that TB03 1430 to 1630 hours
short constant fluctuations of the solar output could be TB04 1630 to 2359 hours
traced via several weather parameters. A review of
literature [3][4] confirm the same. Accordingly, a set of Typical Solar Output Curve
weather parameters are used to forecast the intermittent TB01-30% TB02-50% TB03-15%
fluctuations. 100.00
To select attributes for forecasting, all the weather 80.00
parameters, presented in the weather forecast, were Solar Output(%)
60.00
analyzed with corresponding solar production. From a
40.00
study of data patterns and review of literature [2] [3] [4],
parameters that show a good correlation with solar 20.00
production were extracted. They are cloud cover, 0.00 Time(h)
0:00
1:45
3:30
5:15
7:00
8:45
10:30
12:15
14:00
15:45
17:30
19:15
21:00
22:45
humidity, precipitation, temperature, visibility and gusts.
Correlation among the chosen weather parameters were
tested to eliminate unnecessary complexity of the system. Figure 3. Time Bands
Since Humidity and Temperature show a strong linear As shown in Figure 3, a significant production is not
negative correlation, as depicted by Figure 2, only observed during time bands TB00 and TB04. Therefore,
humidity was selected to be included in the forecasting the proposed system will be active only during the time
model. duration of 0730 to 1630. Energy produced before and
100 Humidity & Temperature Correlation after that would be reserved in the battery.
Humidity(%)

Out of the selected weather attributes, cloud cover,

90 humidity and precipitation were further classified into
weather classes indicated in Figure 4 and Figure 5.
80
Cloud Cover Weather Classes
70 0-10% 6-Sunny
10%-30% 5-Mostly Sunny
Class Ranges

60 Temperature in Celcius
30%-50% 4-Partly Cloudy
25 27 29 31 33
Figure 2. Correlation between Humidity & Temperature 50%-75% 3-Clouds & Sun

Hence, the forecasting model was developed with the five
weather attributes: Cloud cover, Humidity, Precipitation,
Visibility & Gust speed.
2) Data Classification
Data classification was performed under each weather
attribute. Solar and weather data were classified into
several layers according to the month, time of day and
weather classes.
Initial classification was the monthly classification. This
could also have been done based on the seasons. Since Sri
75%-90% 2-Mostly Cloudy
90%-100% 1-Cloudy
Figure 4. Cloud Cover Weather Classes
Six weather classes were identified under the weather
attribute “cloud cover”. Ranges of the weather classes
identified under humidity and precipitation were similar.
The six weather class ranges are depicted along with the
respective weather class numbers in Figure 5.
Summing up, first the acquired weather data are classified
into the selected attributes. Then the solar production
data are fed into all the five weather attributes. Under that,
further classification is carried on as per month, time band
and weather classes. This results in a total of 720 data
classes.
Humidity & Precipitation Weather Classes
Precipitation Weather Classes Humidity Weather Classes
0-10% 1 6
Class Ranges
Forecasting formula for the example is given in equation
(2), where, t is 0.41(time in numerical format) and y is 85.
𝐹𝐶𝐶 (𝑡, 𝑦) = 36.950 + 3.404𝑡 + 5.384𝑦 − 1.962𝑡 2 −
2.264𝑡𝑦 + 2.498𝑦 2 (2)
Data classes under other four weather attributes for the
same scenario could also be determined according to the
respective weather readings. Subsequently, solar
forecasts based on them too could be made.

10%-30% 2 5
30%-50% 3 Aforementioned process of data class identification and
4
50%-75% 4 figuring out the particular formula can be facilitated with
3
75%-85% 5 the aid of decision tree techniques. The outcome of this
2
85%-100% 6 procedure gives five forecasts based on five weather
1
attributes for a given time.
Figure 5. Humidity & Precipitation Weather Classes In addition to the first five forecasts, a combined forecast
3) Formula Generation is also made from the first five forecasts. The formula for
After data classification, time series analysis was the combined forecast is developed using a super
conducted to characterize the data patterns in each data ensemble approach. It involves unequal weighted
class. Time series plot analysis is the major technique used. coefficients acquired by regression analysis. The formula is
Three-dimensional graphs, similar to the graph shown by generated per month, per time band. It is in the format
Figure 6, were plotted with the x-axis as time, y-axis as given by equation (3).
weather attribute (cloud cover, humidity, precipitation, 𝐹𝐶𝑜𝑚 = 𝑎0 + 𝑎1 . 𝐹𝐻 + 𝑎2 . 𝐹𝐶𝐶 + 𝑎3 . 𝐹𝐺 + 𝑎4 . 𝐹𝑉 + 𝑎5 . 𝐹𝑃
visibility, gusts) and z-axis as the solar output. Similar plots where: (3)
were drawn for all data classes under the five weather 𝐹𝑐𝑜𝑚 - Combined model Forecast
attributes. Then, second order polynomials expressing the 𝐹𝐻 - Humidity based Forecast
trend were obtained by smoothening the drawn multiple 𝐹𝐶𝐶 - Cloud Cover based Forecast
time series plots. Quadratic trend analysis is the 𝐹𝐺 - Gust based Forecast
smoothening technique used. The resulting polynomials 𝐹𝑉 - Visibility based Forecast
are of the format given in equation (1). 𝐹𝑃 - Precipitation based Forecast
2
𝐹𝑌 (𝑡, 𝑦) = 𝑃00 + 𝑃10 𝑡 + 𝑃01 𝑦 + 𝑃20 𝑡 + 𝑃11 𝑡𝑦 + 𝑃02 𝑦 2 The coefficients a0, a1, a2, a3, a4 and a5 were obtained
where: (1) for each month separately for the three time bands by
𝐹𝑌 - Solar Forecast based on weather attribute ‘Y’ performing multiple linear regression analysis.
𝑦 - Weather Attribute
B. Forecast Types
𝑡 - Time of Day in numerical format 1) Half Hourly Forecast
Five weather-attribute based forecasts and the combined
forecast can be considered as six sub-models producing six
sub-forecasts under a single forecasting model. This
approach identifies the existing trend prior to a prediction.
It enables the system to follow the weather trends more
closely. Sometimes, for a short time period, the impact of
a single weather parameter may dominate the solar
output than the rest. At another time, a few weather
parameters together may contribute for solar power
fluctuations [5]. Because of this, the availability of six
Figure 6. Model TB02-Weather Class 02 for Humidity forecasts serve to be useful.
Likewise, formulae to predict the solar output for all the Nonetheless, the next question is which forecast should be
data classes, based on each weather attribute, were deemed as the main forecast for a given instance. Thus,
generated. For example, consider the prediction of solar before making the solar forecast for a time step t, mean
output based on the weather attribute cloud cover, for a square errors of previous forecasts (weather attribute-
cloud cover reading of 85% at 0945 hours in May. This based and combined) are calculated. Forecasts of the
scenario belongs to the data class of month May, time three previous time steps (t-1, t-2 and t-3) are considered
band 01 and weather class 02 under the weather attribute for this purpose. The deviations or mean square errors of
cloud cover. As explained, every data class has a formula the predictions from the actual values of the three
to predict the solar output. Thus, by utilizing the previous time steps are obtained. Then the sub-forecast
corresponding formula the forecasting mechanism makes type with the minimum deviation for the last three- time
the prediction of solar output based on cloud cover. steps out of the six is selected. Finally, the prediction for
the tth timestep is made by using the selected sub-forecast
type. This prediction is done thirty minutes in advance.
CEB control center is informed simultaneously.
2) Daily Forecast
Weather parameter-based forecast was brought into the
project because solar irradiation level involved systems,
which failed to trace constant fluctuations of the solar
output [7]. However, as per review of literature [12]
irradiation based methods are more suitable to fulfil the
CEB requirement of daily energy prediction at the dawn of
the day.
Energy prediction in the designed system includes three
components: recent energy levels, level of available
incident solar irradiation and weather factors. Since,
Angstrom linear solar irradiation model [18] has been
identified as the most suitable model to predict the solar
irradiation in Sri Lanka, it is used for the irradiation based
component. After comparison with a few similar models,
Weather Conditioned Moving Average (WCMA) method [7]
was chosen and modified to be used for the previous days-
based energy prediction. For energy prediction based on
weather factors, the combined forecasting, equation (3),
is used. This trio were put together to output a single
energy forecast via weighted coefficients acquired by
multiple linear regression. The relationship between the
three prediction mechanisms are as in equation (4). Three
factors are included to make the result more reliable.
𝐸𝑃 = 𝑏0 + (𝑏1 × 𝐸𝐼𝑟 ) + (𝑏2 × 𝐸𝑃𝑟𝑒𝑣 ) + (𝑏3 × 𝐸𝐶𝑜𝑚 )
where: (4)
𝐸𝑃 -Expected daily energy level
𝐸𝐼𝑟 -Energy prediction based on incident solar
radiation.
𝐸𝑃𝑟𝑒𝑣 -Energy prediction based on previous days.
𝐸𝐶𝑜𝑚 -Energy prediction based on combined weather
forecast.
Review of literature shows that solar irradiation levels
show a considerable variation in the four climate seasons
of Sri Lanka [12]. Hence, relationship between the
prediction mechanisms were developed per season.
Whole forecasting procedure discussed, may be
implemented in software to automatically acquire the
weather data from the accuweather.com and interval
meter data of the solar plant. Then the final prediction for
the corresponding time interval is output and directly
informed the CEB SCC of the prediction.
C. Battery Storage
1) Maintaining the Predicted Output
Predicted power output is maintained by monitoring the
available output. A battery bank and a DC-DC buck-boost
converter is used for this. First, output is sent through a
Buck-Boost converter to ensure that the voltage remains
at a constant level despite of power fluctuations from the
PV cell. The battery bank is charged from the excess of the
forecasted and is discharged across the load when
necessary. For proper operation, the battery bank is
coupled with an automatic switching mechanism to
initiate and stop the charging and discharging processes.
Further, as a strategic contingency option a CEB alert
system is also added to the system to inform the CEB in an
event where the whole system fails to perform as
expected. The alert system monitors the changes in
prediction as well as the instantaneous battery capacity
and checks whether the promised value can be
maintained. It will warn the CEB if the system is unable to
deliver the energy committed.
2) Determining the Battery Capacity
A cost-effective battery storage should neither be too
small nor big. Cost cutting feature of the proposed system
is that battery storage is only used as a supporting limb for
the forecasting. Under the worse-case possible, the
battery storage should be able to supply the peak power
of the solar plant until the SCC is able to transfer the load
to an active plant after CEB alert system is run. This is a
factor to be considered in determining battery capacity.
Considering the operation time of alert system, an hour
has been allocated as the time margin to shut the supply
after the alert. That is, the battery bank should be capable
of supplying the peak power for an hour.
Solar Output (%)
c
Time(h)
a b
Figure 7. Typical Solar Output Curve
When both the morning (a) and evening (b) energy
reservations are also taken into consideration, the
minimum battery capacity required to suffice the
condition of being able to supply the peak power for an
hour was calculated.
III. RESULTS
The forecasting model delivers the solar output forecast at
the beginning of the day via generated daily forecast
formulae for each month utilizing following inputs:
expected solar irradiation, forecasted weather data and
daily energy production of the site of three previous days.
This forecast is informed to the CEB SCC at the beginning
of the day. Along with that, forecasting model also
generate short interval forecasts: half-hourly and quarter
hourly. Half-hourly forecast is utilized to supply solar
power to the national grid in pre-informed half-hourly
steps with the aid of the implemented battery storage and
switching mechanism.
Short interval forecasts are made via generated data class
formulae and logical determination of most suitable
forecast type through sub-model selection. Data class is
identified using fed inputs which are month, time and
weather forecast. The system comprises of six sub-models. Daily energy prediction system is built for the four climate
Five out of the six sub-models consist a multitude of seasons in Sri Lanka. The seasonal factors mostly affect the
formulae to be used under different time bands and incident solar irradiation component of the daily energy
weather conditions. Combined model or the sixth sub- prediction. Figure 9 depicts the predicted solar irradiation
model consists of three formulae of the format given by for the month of May which represents the South-West
equation (3) for the three effective time bands per month. Monsoon. Similar to the short interval forecasting system,
Coefficients generated for the three-time bands for the developed seasonal daily energy prediction system was
month of May are given by Table 2. Most appropriate sub- also tested using the data obtained from test solar site.
model to be used under given circumstances would be Figure 10 presents the performance of the daily energy
logically determined by the forecasting model itself. prediction made on a set of random days during the South-
Table 2. Coefficients of Combined Forecasting Formula for May West Monsoon Season. The coefficient values obtained
for the daily energy model given in equation (4) for the
Coefficient TB 01 TB 02 TB 03
said season were, b0 = -2.45, b1 = -0.24, b2 = 0.49, b3 =
a0 -10.84 -13.34 -5.23
0.83. This model was also validated using the equation (5).
a1 -0.44 1.26 0.03
a2 0.52 0.14 0.71 Resulted rate of accuracy was 93%.
a3 0.41 -0.21 0.56 4.75
May-2018
a4 0.77 -0.04 -0.59
4.7
a5 0.16 0.2 0.61

ENERGY
Submodel selection approach enables the system to adapt 4.65
according to sudden weather variations. Thus, minimizes 4.6
the deviations of predicted and actual solar power output.
Also, the forecasting system includes a contingency plan. 4.55
Quarter-hourly forecasts are used for this. It is being used 4.5 DAYS
by the plant to ready the required back-up to supply the 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
deficit or store the excess. Else, to alert the CEB in the rare Figure 9. Incident Solar Radiation Prediction for the
event of system failing to meet the predicted amount via Month of May
backup.
This forecasting model was validated by testing with the
30
Daily Energy Prediction
solar power data acquired from another site located in Mt.
Lavinia. Figure 8 depicts the results of the forecasting 25
system on a randomly selected day. Rate of accuracy for
20
forecasting was calculated as per the equation (5). It
resulted of an accuracy rate of 88%. 15
10
Forecasting System-13th of May
90 5
Solar Output(%)

80 0
70
May-13th
May-17th
May-18th
May-29th

June-19th
June-04th
June-07th
June-08th
June-15th
June-17th
June-18th

June-20th

June-25th
June-01st

June-21st

60
50
40 Actual Energy Level (kWh) Predicted Energy Level(kWh)
30
Figure 10. Daily Energy Prediction
20
10 The results of the calculation of required battery capacity
0 for the days of test period ranged between 12% to 15% of
the total capacity. That is a 14Ah capacity for the test site.
8:30
9:00
9:30
10:00
10:30
11:00
11:30
12:00
12:30
13:00
13:30
14:00
14:30
15:00
15:30
16:00
16:30

Time (h) IV. FUTURE WORK

15 min-Forecast(%) CEB Forecast(%)-30min Solar panel temperature is recommended to be included
Solar Output (%) as an attribute for forecasting in future work. This is
Figure 8. Short Interval Forecasting System for 13th of identified mainly to track the aftereffects of mid-day cloud
May cover. It was observed during the study that if the clouds
√(𝑃𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 − 𝐴𝑐𝑡𝑢𝑎𝑙)2 cover the solar panel for a short period of time during the
𝑅𝑎𝑡𝑒 𝑜𝑓 𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 (%) = {1 − } × 100%
𝑅𝑎𝑡𝑒𝑑 𝐶𝑎𝑝𝑎𝑐𝑖𝑡𝑦 mid-day, the panel cools down.
(5) This scenario may result in a considerable increment of the
output power after the cloud vanishes. This phenomenon
exhibits that the temperature of the solar panel also has a
significant effect to the power output at such situations.
But the presented forecasting system does not concern
the panel temperature. Therefore, including this attribute
in the forecasting system may increase the rate of
accuracy of the system. Thus, it would be more effective
and accurate to include the panel temperature in future
studies.
V. CONCLUSION
Proposed study forwards a solution for Intermittency of
Solar. Forecasting system and battery storage make it
possible to convert non-dispatchable solar power to a
semi-dispatchable power source. The forecasting system
predicts the solar power output based on solar irradiance
and five weather attributes: humidity, cloud cover,
precipitation, visibility, gust. Suitable battery storage aids
in maintaining the forecasted power output. Semi-
dispatchability is achieved by providing power to the
utility provider in a series of steps with pre-defined power
outputs.
Results show acceptable accuracy rates of 93% and 88%
for daily energy predictions and half-hourly predictions
respectively. The system reliability is further improved by
adding the battery which amounts to 15% of the installed
capacity. In the eventuality that the system is unable to
supply scheduled power, the system includes an alert
system for precautionary actions.
The improvements aid the utility service provider, as it not
only reduces the burden on the system, but also reduces
the regularity of quick response power stations burning
expensive fuel to compensate for the lack of consistency.
The value-added service of improving the power injection
to the system from solar qualifies for an increase in
payment from CEB for power from solar, as well as an
increase in the allowance for solar PV from private
investors. Thus, the study proposes a techno-economic
solution for semi-dispatchable solar.
Presented concept and initial framework could be carried
forward in accomplishing the ultimate objective of
increasing the renewable energy integration in the
national grid.
VI. REFERENCES
[1] CEB, "Long Term Generation Expansion Plan (2018-
2037)," PUCSL, 2018.
[2] M. Detyniecki, "Weather-based solar energy
prediction," 2012 IEEE International Conference on
Fuzzy Systems, 2012.
[3] W. D. A. S. Wijayapala and D. H. K. Kushal,
"Development of a model for prediction of solar
radiation," Engineer: Journal of the Institution of
Engineers, Sri Lanka, vol. 48, no. 3, p. 19, 2015.
[4] A. W. Zhu and H. Pi, "A Method for Improving the
Accuracy of Weather Forecasts Based on a
Comprehensive Statistical Analysis of Historical Data for
the Contiguous United States," Journal of Climatology &
Weather Forecasting, vol. 2, no. 1, 2014.
[5] X. Shao, S. Lu and H. F. Hamann, "Solar radiation
forecast with machine learning," 2016 23rd
International Workshop on Active-Matrix Flatpanel
Displays and Devices (AM-FPD), 2016.
[6] S. A. Fatemi and A. Kuh, "Solar radiation forecasting
using zenith angle," 2013 IEEE Global Conference on
Signal and Information Processing, 2013.
[7] Z. Jiang, X. Jin and Y. Zhang, "A Weather-Condition
Prediction Algorithm for Solar-Powered Wireless Sensor
Nodes," 2010 International Conference on
Computational Intelligence and Software Engineering,
2010.
[8] "Department of Meteorology," [Online]. Available:
http://meteo.gov.lk/index.php?lang=en. [Accessed 28
10 2018].
[9] "Accuweather," Accuweather.inc, [Online]. Available:
https://www.accuweather.com/en/lk/mount-
lavinia/311485/may-
weather/311485?monyr=5/1/2018&view=table.
[Accessed 28 10 2018].
[10] R. Kumar, "Decision Tree for the Weather Forecasting,"
International Journal of Computer Applications, vol. 76,
2013.
[11] S. Ghosh, S. Rahman and M. Pipattanasomporn,
"Distribution Voltage Regulation Through Active Power
Curtailment With PV Inverters and Solar Generation
Forecasts," IEEE Transactions on Sustainable Energy,
vol. 8, no. 1, pp. 13-22, 2017.
[12] S. Sendanayake, N. P. Miguntanna and M. Jayasinghe ,
"Estimating Incident Solar Radiation in Tropical Islands
with Short Term Weather Data," European Scientific
Journal, vol. 10, no. 1857-7881, 2014.
[13] J. Marcos, I. d. l. Parra, M. Garcia and L. Marroyo,
"Control Strategies to Smooth Short-Term Power
Fluctuations in Large Photovoltaic Plants Using Battery
Storage Systems," Energies 2014, no. 1996-1073, 2014.
[14] G. Koeppel and M. Korpas, "Using storage devices for
compensating uncertainties caused by non-
dispatchable generators," 2006 International
Conference on Probabilistic Methods Applied to Power
Systems, 2006.
[15] S. Podder and M. Z. R. Khan, "Comparison of lead acid
and Li-ion battery in solar home system of Bangladesh,"
2016 5th International Conference on Informatics,
Electronics and Vision (ICIEV), 2016.
[16] D. Chaturvedi and Isha, "Solar Power Forecasting: A
Review," International Journal of Computer
Applications, vol. 145, no. (0975 – 8887), 2016.
[17] M. Anvari, G. Lohmann and M. Wächter, "Short term
fluctuations of wind and solar power systems," New
Journal of Physics, vol. 18, no. 06, p. 063027, 2016.
[18] C. Wan, J. Zhao and Y. Song, "Photovoltaic and solar
power forecasting for smart grid energy management,"
CSEE Journal of Power and Energy Systems, vol. 1, no. 4,
pp. 38-46, 2015.